DARTS: Differentiable Architecture Search
ICLR
Hanxiao Liu, Karen Simonyan, Yiming Yang
This paper addresses the scalability challenge of architecture search by formulating the task in a differentiable manner. Unlike conventional approaches of applying evolution or reinforcement learning over a discrete and non-differentiable search space, our method is based on the continuous relaxation of the architecture representation, allowing efficient search of the architecture using gradient descent. Extensive experiments on CIFAR-10, ImageNet, Penn Treebank and WikiText-2 show that our algorithm excels in discovering high-performance convolutional architectures for image classification and recurrent architectures for language modeling, while being orders of magnitude faster than state-of-the-art non-differentiable techniques. Our implementation has been made publicly available to facilitate further research on efficient architecture search algorithms.
@article{DBLP:journals/corr/abs-1806-09055, author = {Hanxiao Liu and Karen Simonyan and Yiming Yang}, title = {{DARTS:} Differentiable Architecture Search}, journal = {CoRR}, volume = {abs/1806.09055}, year = {2018}, url = {http://arxiv.org/abs/1806.09055}, archivePrefix = {arXiv}, eprint = {1806.09055}, timestamp = {Mon, 13 Aug 2018 16:49:10 +0200}, biburl = {https://dblp.org/rec/journals/corr/abs-1806-09055.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} }